Data Scientist

Actify Data Labs

Bangalore, India

2020 - 2018

Project: Crop Classification using Satellite Imagery

• Business problem: Classification of various crops using Satellite Imagery
• Solution: Built a pipeline to create spectral indices using satellite images, create features on top of the spectral indices, train models and score the models on new satellite images
• Benefit: Reduced huge number of man-hours taken to identify and geospatial labeling of crops across various regions in the country. Manual geolabelling of crops for each district takes around 4-5 man hours. Reduced it down to less than 1 hour without compromising precision and recall

Product Engineering and Management: Interactive Decision Tree

• Developed an interactive drag and drop decision tree tool for segmentation of data for business users. The features include delete, add, cut and copy a subtree from the tree and export the rules to segment the data
• Conceptualized, designed and developed the complete product module including algorithm design, UI design and scalibilty for larger data
• The product is sold to one of the cab sharing unicorn in the country

Project: Estimation of salinity and depth of ponds using Satellite Imagery

• Business problem: Predicting salinity and depth of ponds using Satellite Imagery
• Solution: Built pipeline to automatically download satellite images, process the images to extract structured data for the ponds and deployed machine learning models (xgboost and RF) to predict salinity and depth of the ponds
• Benefit: The salinity is measured manually for 90+ ponds where each measurement takes 15 min and involves travelling 40+ km around the perimeter of ponds. The solution helped in saving 1000s of man hours and accurately predicting the salinity

Project: Identify fraudulent companies using annual filing data for the PMO’s SFIO

• Business problem: Identify fraudulent companies using the annual filing data
• Solution: Development of interpretable gradient boosted machines to identify shell companies and using the same to augment human intelligence
• Benefit: The solution helped investigators identify potential suspect shell companies with a minimal false positive. It was also able to identify more than 95% of all cases that were identified by human intelligence. The solution also provided transparency into the reasons of identifying a potential suspect, this helped investigators in pin-pointing documents for further investigation

Analytics Specialist

Opera Solutions

Delhi, India

2018 - 2017

Project: Revenue Management

• Problem: Identification of revenue leakage during billing for US healthcare providers.
• Solution: Development of an ensemble of Random Forest and Logistic Regression to identify potential revenue items that are missing from a given bill and provide recommendations to the front line accounts.
• Benefit: More than 85 % of the recommendations were correct in identifying missed bill items. This led to significant revenue enchancement for the health care providers.

Analyst

Cognizant Technology Solutions

Chennai, India

2017 - 2016

Project: Ad-Viewership Analytics

• Problem: Estimation of ad spend effectiveness for a major US telecom giant.
• Solution: Built data pipelines to curate, consolidate and process ad-viewership data aross all viewing channels
• Tools used are Unix Scripting, Hadoop, SQL, Hive and Spark

Bachelor’s and Master’s

IIT Bombay

Bombay, India

2016 - 2011

• Energy Science and Engineering. GPA 8.04/10